© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology TECHNICAL REPORT A genomics-guided approach for discovering and expressing cryptic metabolic pathways Emmanuel Zazopoulos1, Kexue Huang1, Alfredo Staffa1, Wen Liu2, Brian O. Bachmann1, Koichi Nonaka2, Joachim Ahlert2,3, Jon S. Thorson2,3, Ben Shen2,4, and Chris M. Farnet1* Published online 21 January 2003; doi:10.1038/nbt784 Genome analysis of actinomycetes has revealed the presence of numerous cryptic gene clusters encoding putative natural products1,2. These loci remain dormant until appropriate chemical or physical signals induce their expression. Here we demonstrate the use of a high-throughput genome scanning method to detect and analyze gene clusters involved in natural-product biosynthesis. This method was applied to uncover biosynthetic pathways encoding enediyne antitumor antibiotics in a variety of actinomycetes. Comparative analysis of five biosynthetic loci representative of the major structural classes of enediynes reveals the presence of a conserved cassette of five genes that includes a novel family of polyketide synthase (PKS)3,4. The enediyne PKS (PKSE) is proposed to be involved in the formation of the highly reactive chromophore ring structure (or “warhead”) found in all enediynes3,4. Genome scanning analysis indicates that the enediyne warhead cassette is widely dispersed among actinomycetes. We show that selective growth conditions can induce the expression of these loci, suggesting that the range of enediyne natural products may be much greater than previously thought. This technology can be used to increase the scope and diversity of natural-product discovery. We have developed a high-throughput genome scanning method to discover metabolic loci independently of their expression. This approach takes advantage of the fact that the genes required for secondary metabolite biosynthesis are typically clustered together in a bacterial genome5. A shotgun DNA sequencing approach is used to generate short (700 bp) random genome sequence tags (GSTs) from a library of genomic DNA prepared from a microorganism. GSTs derived from genes that are likely to be involved in the biosynthesis of natural products are identified by sequence comparisons to a database of microbial gene clusters known to be involved in natural-product biosynthesis. Selected GSTs are then used to design screening probes to identify cloned subgenomic fragments (for example, cosmids or bacterial artificial chromosomes (BACs)) containing the genes of interest as well as the neighboring genes that together may constitute a biosynthetic gene cluster (Fig. 1). Genome scanning provides an efficient way 1Ecopia BioSciences, Inc., 7290 Frederick Banting, Montreal, Quebec H4S 2A1, Canada. 2Division of Pharmaceutical Sciences, 3Laboratory for Biosynthetic Chemistry, and 4Department of Chemistry, University of Wisconsin, Madison, WI 53706. *Corresponding author (farnet@ecopiabio.com). www.nature.com/naturebiotechnology • to discover natural-product gene clusters because the analysis of a relatively small number of GSTs provides reasonable assurance of full genome representation. For example, analysis of 1,000 GSTs from a genome of 8.5 Mbp (the approximate size of the genome of an antibiotic-producing actinomycete) provides DNA sequence sampling every 8.5 kbp (assuming random library coverage). Given that natural-product gene clusters range in size from 20 to 200 kbp6,7 (C.F., unpublished data), it is expected that any given gene cluster will be represented by anywhere from 2 to more than 20 of 1,000 GSTs analyzed. To date, we have used the genome scanning approach to successfully identify more than 450 naturalproduct gene clusters in a variety of actinomycetes (C.F., unpublished data). We used the genome scanning method to isolate enediyne biosynthesis genes from a variety of actinomycete strains known to produce enediynes, a potent class of antitumor antibiotics8. The enediynes induce irreversible DNA damage by a mechanism that involves cycloaromatization of the warhead chromophore (Fig. 2A) to form highly reactive benzenoid diradicals that strip hydrogen atoms from the sugar phosphate backbone of the DNA helix9. We chose the dynemicin and macromomycin biosynthesis gene clusters to demonstrate the effectiveness of the genome scanning method. Comparison with the C-1027 (ref. 3), calicheamicin4, and neocarzinostatin (W. L., K.N., L. Nie, J. Bae, and B.S., unpublished data) gene clusters reveals that the homology among all these loci is limited to a set of five genes, including the gene encoding PKSE, that form a putative “warhead gene cassette” (Fig. 2B). The conserved genes are generally arranged in a presumed operon with Figure 1. A diagrammatic view of the genome scanning method for high-throughput discovery of natural-product biosynthetic gene clusters. Natural-product biosynthetic genes (in color) are clustered in the bacterial genome (for simplicity, only a single gene cluster is shown). High-molecular-weight genomic DNA is randomly fragmented and small fragments are used to prepare a genome sampling library (GSL) in a plasmid vector while large fragments are used to prepare a cluster identification library (CIL) in a cosmid or BAC vector. Gene sequence tags (GSTs) are generated from the GSL clones using a universal primer located in the plasmid vector. The GSTs are compared to a database of natural-product biosynthetic genes to identify tags derived from genes involved in natural-product biosynthesis (“hot” GSTs, colored inserts; step 1). These genes are then used as probes to identify CIL clones containing the corresponding genes as well as their neighboring genes (“hot” CIL clones). Overlapping CIL clones may be identified by restriction fragment length mapping or during the subsequent sequencing step. The hot CIL clones are randomly fragmented and used to prepare a second plasmid library that provides templates for sequencing (step 2). Sequencing and assembly of the selected CIL clones result in a complete natural-product gene cluster that is then annotated and entered into the database (step 3). FEBRUARY 2003 • VOLUME 21 • nature biotechnology 187 TECHNICAL REPORT A D We compared the other warhead cassette proteins to protein 007A sequences present in the GenBank nonredundant data009C base to assess putative functions. 2 R , R = sugars 1 028D One protein family (TEBC) is similar to the 4-hydroxybenzoyl054A CoA thioesterase of Pseudomonas sp. strain CBS-3 in regions of the 059A protein that have been shown to 132H have an important role in catalysis11 and thus may be involved in 135E polyketide chain release, cycliza145B tion, or both (see Supplementary 3 4 Fig. 3 online). Three families of E B unknown proteins (UNBL, 046E DYNE UNBV, and UNBU) show no significant homology to proteins in CALI 100B the public databases and therefore represent novel protein famMACR 171B ilies that appear to be specific to NEOC enediyne biosynthetic loci. Structural analysis of the UNBV C-1027 PKSE TEBC UNBL UNBV UNBU UNBU proteins predicts that they are secreted proteins with N-termiC ACP nal signal sequences, whereas the NH KS ? KR DH ? PPTE COOH AT UNBU proteins are predicted to be integral membrane proteins with seven or eight putative Figure 2. Chemical structures of enediynes and genes involved in warhead formation. (A) The structures of membrane-spanning alpha the enediynes dynemicin (1), calicheamicin (2), neocarzinostatin (3), and C-1027 (4). The common cyclododecylpolyene skeleton found in all warheads is highlighted in red. The complete structure of helices (see Supplementary Figs. macromomycin has yet to be elucidated; however, the limited structural information available is consistent 4–6 online). Although the funcwith a chromophore ring system similar to that found in C-1027 (ref. 20). (B) Organization of the warhead tions of the TEBC, UNBL, UNBV, gene cassettes found in the dynemicin (DYNE), calicheamicin (CALI), macromomycin (MACR), and UNBU proteins remain neocarzinostatin (NEOC), and C-1027 loci. (C) Domain organization of the warhead PKS, consisting of KS (ketoacyl synthase), AT (acyl transferase), ACP (acyl carrier protein), KR (keto reductase), DH unknown, their strict association (dehydratase), and PPTE (4′-phosphopantetheinyl transferase). (D) Organization of the warhead cassette with the warhead PKS and their genes found in loci from actinomycete strains not previously reported to produce enediyne natural products. presence in all enediyne biosyn(E) Warhead cassette genes from actinomycete strains newly isolated from soil samples. 007A, locus found thetic loci strongly suggest that in Amycolatopsis orientalis; 009C, locus found in Streptomyces ghanaensis; 028D, locus found in Kitasatosporia sp.; 054A, locus found in Micromonospora megalomicea subsp. nigra; 059A, locus found in they have essential roles in the Streptomyces cavourensis subsp. washingtonensis; 132H, locus found in Saccharothrix aerocolonigenes; formation, stabilization, or trans135E, locus found in Streptomyces kaniharaensis; 145B, locus found in Streptomyces citricolor. Loci 046E, port of the enediyne warhead. 100B, and 171B were found in new actinomycete isolates (Ecopia BioSciences Inc.). We used the genome scanning method to isolate natural-produnidirectional transcription and frequent overlap of translational uct biosynthetic loci from a variety of actinomycete strains that start and stop codons, suggesting that their gene products are have been reported to produce various classes of natural products coordinately expressed and functionally related. As these are the but not enediyne compounds. Out of 50 actinomycete strains anaonly genes common to all enediyne loci analyzed to date, we prolyzed, eight (16%) were found to contain biosynthetic loci conpose that they constitute a functional unit responsible for the biotaining the enediyne warhead cassette (Fig. 2D), indicating that genesis of the warhead, the single structural feature that is found these strains could potentially produce enediyne natural products. in all of the known enediynes9 (Fig. 2A). This finding is surprising, as none of the eight strains was previThe PKSEs are likely to generate the carbon skeleton of the warously reported to produce enediynes, and it indicates that head by catalyzing iterative cycles of acyl-coenzyme A (acyl-CoA) enediyne biosynthetic loci occur at an unexpectedly high frequencondensation, ketoreduction and dehydration, using an acyl carricy in the actinomycetes. Enediyne loci occurred at a similar freer protein (ACP) domain as a covalent attachment site for the quency in actinomycete strains newly isolated from soil samples: 3 growing carbon chain. The PKSEs contain enzymatic domains out of 20 (15%) randomly selected strains were found to harbor characteristic of known PKSs, including ketoacyl synthase (KS), biosynthetic loci containing enediyne warhead cassette genes (Fig. acyltransferase (AT), ketoreductase (KR), and dehydratase (DH) 2E). It is notable that the strong conservation of the enediyne wardomains, as well as ACP domains3,4 (Fig. 2C and Supplementary head gene sequence and gene order holds across several actinoFig. 1 online). Additional analysis of the PKSE sequences mycete genera (Actinomadura, Amycolatopsis, Kitasatosporia, described here further revealed a previously unidentified domain Micromonospora, Saccharothrix, Streptomyces). in the C-terminal region of the protein that is similar to 4′-phosFinally, we validated enediyne production by culturing strains phopantetheinyl transferases10 (PPTases) and is likely to be harboring warhead gene cassettes in a variety of fermentation involved in post-translational autoactivation of the PKSE (Fig. 2C media to test for activity in the biochemical induction assay (BIA), and Supplementary Figs. 1, 2 online). a modified prophage induction assay that detects agents that damHO O HO H3CO SSSCH3 OH H N R 2O OH O O NH O O H N H3C HO OR1 S O CH3 O CH3 I O CH3 H3COOC HOOC OCH3 O 1 2 CH2 O H3CO © 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology N H O O O CH3 O O O O O O H3C O O OH O H3C O CH3HN HO O O (H3C)2N HO O O CH3 OH OH Cl NH2 CH3 OH 2 188 nature biotechnology • VOLUME 21 • FEBRUARY 2003 • www.nature.com/naturebiotechnology TECHNICAL REPORT © 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology A cates that additional classes of enediyne natural products remain to be discovered. The genome sequences of two actinomycetes, Streptomyces coelicolor1 and Streptomyces avermitilis2, reveal several sets of apparently cryptic natural-product gene clusters, suggesting that these well-studied strains may produce a greater number of bioactive compounds than has been detected by fermentation broth analysis16. We have demonstrated here that standard fermentation broth screening failed to identify many bacterial strains that can produce enediyne antibiotics. The ability of actinomycetes to produce antibiotics and other bioactive natural products has apparently been greatly underestimated. This work demonstrates the utility of genome analysis in discovering hidden metabolic potential and directing rational approaches for the expression, detection, and purification of new bioactive natural products. B Experimental protocol Figure 3. Enediyne production measured by the BIA. The presence of inhibition zones surrounded by blue rings is indicative of β-galactosidase induction due to DNA damage. At higher dilutions, only β-galactosidase activity is observed, as the DNA damaging activity of the enediynes occurs at concentrations that are not bactericidal12. (A) BIA activity from microorganisms producing calicheamicin, dynemicin, macromomycin, and neocarzinostatin and (B) from Amycolatopsis orientalis (007), Streptomyces ghanaensis (009), and Streptomyces citricolor (145), organisms not previously reported to produce enediynes, as well as from two new actinomycete isolates, 046 and 171. Media AA, YA, and ZA do not support enediyne production. age DNA and is commonly used to assay enediyne production12. In most media these strains did not have detectable BIA activity. However, all strains produced BIA activity when grown in specialized media selected for their ability to support enediyne production (Fig. 3). These results provide strong evidence that these strains are able to produce enediyne natural products and that the expression of enediyne biosynthetic loci is restricted to certain fermentation conditions. Although all of the 11 loci encoding unknown enediynes contain the conserved warhead cassette genes, these loci differ considerably from one another. They also differ from the loci that encode the structurally characterized enediynes in the surrounding genes, which probably generate the structural units appended to the warhead chromophore (unpublished data). This may indicate that each locus potentially encodes a different enediyne natural product. Considering that a total of only 11 members of the enediyne family have been described to date9, it is likely that some of the unknown loci described here produce new classes of enediynes. As the enediynes are the most potent antitumor agents ever discovered8, the discovery of new classes holds great interest. While they are too toxic for systemic use in unmodified form, the enediynes have proven to be effective anticancer drugs when conjugated to polymers or antibodies. For example, a polymer-conjugated form of neocarzinostatin has been used clinically to treat hepatoma in Japan since 199413, whereas a calicheamicin–antiCD33 antibody conjugate (Mylotarg) was approved in the United States in 2000 for the treatment of acute myelogenous leukemia14. In addition, several C-1027–antibody conjugates are currently under clinical evaluation as anticancer drugs15. This work indiwww.nature.com/naturebiotechnology • Genome scanning. The genomes of the dynemicin-producing organism (Micromonospora chersina strain M956-1, ATCC 53710) and the macromomycin-producing organism (Streptomyces macromyceticus strain M480M1, NRRL B-5335) were analyzed by genome scanning (described in patent application CA 2,352,451). Briefly, high-molecular-weight genomic DNA was prepared from each organism17 and used to generate a smallinsert genomic sampling library (GSL) and a large-insert cluster identification library (CIL). Both libraries contain randomly fragmented genomic DNA and therefore are representative of the entire genome. For the generation of the GSL, genomic DNA was sonicated and fragments of 1.5–3 kbp were prepared by agarose gel electrophoresis and cloned into plasmid vectors. For the generation of the CIL, genomic DNA was fragmented to a size range of 30–50 kbp by partial digestion with the restriction endonuclease Sau3A1 before being cloned into cosmid vectors. One thousand gene sequence tags (GSTs) (average read length, 700 bp) were obtained from each GSL, translated into amino acid sequence, and compared to a proprietary database of microbial natural-product biosynthetic loci (DECIPHER Database, Ecopia BioSciences Inc., Montreal, Canada; http://www.ecopiabio.com) using the basic local alignment search tool protein database (BLASTP) software (http://www.ncbi.nlm.nih.gov/) to identify gene sequences likely to be involved in the production of natural products. The efficiency of the genome scanning method depends in part on the ability to distinguish genes involved in natural-product biosynthesis from those involved in primary metabolism, and thus will vary according to the size and breadth of the database used for comparison. The probability that a particular gene cluster will be identified by analysis of a given number of GSTs is improved if the database contains a large number of gene clusters representing a broad range of natural-product classes. The DECIPHER database was initially populated with gene clusters representing a diverse range of natural product classes collected from public databases such as GenBank, and subsequently enriched with gene clusters discovered at Ecopia. Selected gene sequences were used to design screening probes to identify cosmids containing putative natural-product gene clusters from the CIL. Selected cosmids were sequenced by the shotgun method, and overlapping cosmids were identified by using the cosmid end sequences as probes to screen the CILs. Genome scanning was also used to isolate natural-product biosynthetic loci from 50 previously isolated actinomycete strains as well as from 20 new actinomycete strains isolated from soil samples. Actinomycete strains used to isolate natural-product biosynthetic loci include Amycolatopsis orientalis ATCC 43491 (vancomycin producer), Streptomyces ghanaensis NRRL B-12104 (moenomycin producer), Kitasatosporia sp. CECT 4991 (taxane producer), Micromonospora megalomicea subsp. nigra NRRL 3275 (megalomicin producer), Streptomyces cavourensis subsp. washingtonensis NRRL B-8030 (chromomycin producer), Saccharothrix aerocolonigenes ATCC 39243 (rebeccamycin producer), Streptomyces kaniharaensis ATCC 21070 (coformycin producer), Streptomyces citricolor IFO 13005 (aristeromycin and neplanocin A producer). Enediyne biosynthetic loci were identified by the presence of the conserved enediyne warhead cassette genes as well as other genes frequently found in biosynthetic loci encoding other natural-product classes (data not shown). The neocarzinostatin FEBRUARY 2003 • VOLUME 21 • nature biotechnology 189 TECHNICAL REPORT © 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology locus was cloned from Streptomyces carzinostaticus subsp. neocarzinostaticus ATCC 15944, sequenced by the shotgun method, and confirmed to direct neocarzinostatin biosynthesis by gene inactivation and complementation experiments (W. L., K.N., L. Nie, J. Bae, and B.S., unpublished data). Protein homology analysis. To identify the PKSE PPTase domain, the Cterminal regions of the PKSEs from the neocarzinostatin, calicheamicin and macromomycin biosynthetic loci were analyzed for their folding using secondary structure predictions and solvation potential information18. Comparison searches using a database of known three-dimensional structures of proteins revealed similarities with Sfp, the 4′-phosphopantetheinyl transferase from the Bacillus subtilis surfactin biosynthetic locus19 (PDB id: 1QR0). Protein alignments based on secondary structure predictions as well as identification of conserved amino acids important for cofactor binding can be found in the Supplementary Figure 2 online. Amino acid sequence alignments of the PKSE, TEBC, UNBL, UNBV, and UNBU proteins from the calicheamicin, macromomycin, dynemicin, C-1027, and neocarzinostatin biosynthetic loci can also be found in the Supplementary Figures 1 and 3–6. Where applicable, putative functions for these proteins were assessed by comparison to protein sequences present in the GenBank nonredundant database using the BLASTP software and by subcellular protein localization prediction using the PSORT program available at http://psort.nibb.ac.jp./ Fermentation and activity screening. Organisms were initially grown in 25 ml of TSB17 seed medium for 60 h at 28 °C and then diluted 30-fold in 25 ml production medium. Production medium for calicheamicin was composed of 20 g of sucrose, 2 g of Bactopeptone (Becton Dickenson, Sparks, MD), 5 g of cane molasses, 0.1 g of FeSO4•7H2O, 0.2 g of MgSO4•7H2O, 0.5 g of KI, and 5 g of CaCO3 per liter. Production medium for macromomycin was composed of 40 g of glucose, 5 g of dried yeast, 1 g of K2HPO4, 1 g of MgSO4, 1 g of NaCl, 2 g of (NH4)2SO4, 2 g of CaCO3, 1 mg of FeSO4•7H2O, 1 mg of MnCl2•4H2O, 1 mg of ZnSO4•7H2O, and 0.5 mg of NaI per liter. Production medium for dynemicin was composed of 10 g of corn starch, 5 g of Pharmamedia (Southern Cotton Oil Co., Memphis, TN), 1 g of CaCO3, 0.05 g of CuSO4•5H2O, and 0.5 mg of NaI per liter. Production medium for neocarzinostatin was composed of 40 g of glucose, 15 g of casamino acids, 5 g of NaCl, 2 g of CaCO3, 1 g of K2HPO4 and 12.5 g of MgSO4 per liter. Selective media supporting enediyne production in organisms not previously reported to express enediyne compounds and new actinomycete isolates were as follows. For A. orientalis (007), medium was as described for calicheamicin production. For S. ghanaensis (009), production medium was composed of 30 g of glycerol, 15 g of distiller’s solubles, 10 g of Pharmamedia, 10 g of fish meal, and 6 g of CaCO3 per liter. For S. aerocolonigenes (132), S. kaniharaensis (135), and Ecopia strain 171, production media were composed of 60 g of molasses, 20 g of soluble starch, 20 g of fish meal, 0.1 g of CuSO4•5H2O, 0.5 mg of NaI, and 2 g of CaCO3 per liter. For S. citricolor (145) and Ecopia strain 046, the production medium was composed of 10 g of glucose, 10 g of starch, 15 g of soybean meal, 1 g of K2HPO4, 3 g of NaCl, 1 g of MgSO4•7H2O, 7 mg of CuSO4•5H2O, 1 mg of FeSO4•7H2O, 8 mg of MnCl2•4H2O, and 2 mg of ZnSO4•5H2O per liter. For S. cavourensis subsp. washingtonensis (059), production medium was composed of 20 g of glucose, 5 g of Bactopeptone, 5 g of beef extract, 5 g of NaCl, 3 g of yeast extract, and 2 g of CaCO3 per liter. Examples of media not supporting enediyne production include media AA (10 g of glucose, 40 g of corn dextrin, 15 g of sucrose, 10 g of casein hydrolysate, 1 g of MgSO4•7H2O, and 2 g of CaCO3 per liter) and CECT media 32 and 131 (Colección Española de Cultivos Tipo, Valencia, Spain) herein referred to as media YA and ZA, respectively. Production cultures (25 ml) were incubated for 7 d at 28 °C under constant agitation. Culture (2 ml) was removed and clarified by centrifugation to provide supernatant samples. The rest of the culture (supernatant and mycelia) was extracted with an equal volume of methanol under agitation for 30 min. Extracts were clarified by centrifugation and diluted accordingly in their respective media supplemented with 50% methanol. The BIA was performed as described12. Briefly, 10 µl of supernatant or extract and twofold serial dilutions thereof were applied to agar plates seeded with Escherichia coli BR513 and incubated for 3 h at 37 °C. Soft 190 nature biotechnology • VOLUME 21 • agar containing 0.7 mg/ml of X-gal was added onto the plate and color development was observed within 30 min. BIA activity was tested on serial dilutions of methanol extracts for microorganisms producing calicheamicin and dynemicin as well as for microorganisms 046, 145, and 171. Serial dilutions of supernatants from microorganisms producing macromomycin and neocarzinostatin as well as from microorganisms 007 and 009 were assayed for BIA activity. All production media used in this study were assayed alone and shown to be negative for BIA activity (data not shown). GenBank accession numbers. The DNA and protein sequences described here are deposited in GenBank under accession numbers AF548580, AF548581, and AF546139–AF546157. Note: Supplementary Biotechnology website. information is available on the Nature Acknowledgments We thank S. Mercure, V. Dodelet, and M. Piraee for helpful discussions and J. McAlpine for critical reading of the manuscript. B.S. is a recipient of a NSF CAREER Award (MCB9733938) and a NIH Independent Scientist Award (AI51689). Enediyne studies in the Shen lab are supported in part by NIH grant CA78747. Research in the Thorson lab is supported in part by NIH grants CA84347, GM58196, and AI52218. J.S.T. is an Alfred P. Sloan Fellow. Competing interests statement The authors declare that they have no competing financial interests. Received 21 October 2002; accepted 4 December 2002 1. Bentley, S.D. et al. Complete genome sequence of the model actinomycetes Streptomyces coelicolor A3(2). Nature 417, 141–147 (2002). 2. Omura, S. et al. Genome sequence of an industrial microorganism Streptomyces avermitilis: deducing the ability of producing secondary metabolites. Proc. Natl. Acad. Sci. USA 98, 12215–12220 (2001). 3. Liu, W. Christenson, S.D., Standage, S. & Shen, B. Biosynthesis of the enediyne antitumor antibiotic C-1027. Science 297, 1170–1173 (2002). 4. Ahlert, J. et al. The calicheamicin gene cluster and its iterative type I enediyne PKS. Science 297, 1173–1176 (2002). 5. Martin, J.F. & Liras, P. Organization and expression of genes involved in the biosynthesis of antibiotics and other secondary metabolites. Annu. Rev. Microbiol. 43, 173–206 (1989). 6. Beyer, S., Distler, J. & Piepersberg, W. The str gene cluster for the biosynthesis of 5′-hydroxystreptomycin in Streptomyces glaucescens GLA.0 (ETH 22794): new operons and evidence for pathway-specific regulation by StrR. Mol. Gen. Genet. 250, 775–784 (1996). 7. Schwecke, T. et al. The biosynthetic gene cluster for the polyketide immunosuppressant rapamycin. Proc. Natl. Acad. Sci. USA 92, 7839–7843 (1995). 8. Doyle, T.W. & Borders, D.B. Enediyne antitumor antibiotics. in Enediyne Antibiotics as Antitumor Agents (eds. Borders, D.B. & Doyle, T.W.) 1–15 (Marcel Dekker, New York, 1995). 9. Smith, A.L. & Nicolaou, K.C. The enediyne antibiotics. J. Med. Chem. 39, 2103–2117 (1996). 10. Walsh, C.T., Gehring, A.M., Weinreb, P.H., Quadri, L.E. & Flugel, R.S. Posttranslational modification of polyketide and nonribosomal peptide synthases. Curr. Opin. Chem. Biol. 1, 309–315 (1997). 11. Benning, M.M. et al. The three-dimensional structure of 4-hydroxybenzoyl-CoA thioesterase from Pseudomonas sp. Strain CBS-3. J. Biol. Chem. 273, 33572–33579 (1998). 12. Elespuru, R.K. & Yarmolinsky, M.B. A colorimetric assay of lysogenic induction designed for screening potential carcinogenic and carcinostatic agents. Environ. Mutagen. 1, 65–78 (1979). 13. Maeda, H. The clinical effects of neocarzinostatin and its polymer conjugate, SMANCS. in Enediyne Antibiotics as Antitumor Agents (eds. Borders, D.B. & Doyle, T.W.) 363–380 (Marcel Dekker, New York, 1995). 14. Sievers, E.L. & Lineberger M. Mylotarg: antibody-targeted chemotherapy comes of age. Curr. Opin. Oncol. 13, 522–527 (2001). 15. Brukner, I. C-1027 Taiho Pharmaceutical Co. Ltd. Curr. Opinion Oncologic, Endocrine & Met. Invest. Drugs 2, 344–352 (2000). 16. Chater, K.F. & Bibb, M.J. Regulation of bacterial antibiotic production. In Biotechnology vol. 7: Products of Secondary Metabolism (eds. Kleinkauf, H. & von Döhren, H.) 57–105 (VCH Press, Weinheim, Germany, 1997). 17. Kieser, T., Bibb, M.J., Buttner, M.J., Chater, K.F. & Hopwood, D.A. Practical Streptomyces Genetics (The John Innes Foundation, Norwich, UK, 2000). 18. Kelley, L.A., MacCallum, R.M. & Sternberg M.J. Enhanced genome annotation using structural profiles in the program 3D-PSSM. J. Mol. Biol. 299, 499–520 (2000). 19. Reuter, K., Mofid, M.R., Marahiel, M.A. & Ficner, R. Crystal structure of the surfactin synthetase-activating enzyme Sfp: a prototype of the 4′-phosphopantetheinyl transferase superfamily. EMBO 18, 6823–6831 (1999). 20. Kumada, Y. et al. A degradation product of the chromophore of auromomycin. J. Antibiot. (Tokyo) 36, 200–202 (1983). FEBRUARY 2003 • www.nature.com/naturebiotechnology